Model-Free Quality Optimization Strategy for a Batch Process with Short Cycle Time and Low Operational Cost

Abstract

Batch processes are an important factor in modern industries. Quality optimization of batch processes to determine the proper settings of key process variables is critical to successful production. The traditional method for quality optimization is model-based optimization (MBO), which is a challenging task in many cases. The accuracy of the model can deteriorate, particularly when process conditions are frequently changed, a situation that commonly occurs in some batch processes. In this study, a systematic, model-free optimization (MFO) strategy is proposed for a batch process with short cycle time and low operational cost. Instead of developing a model to correlate the relationship between the process and its quality variables, MFO optimizes the process on the basis of online experimentation rather than function evaluation. A direct search algorithm is proposed by integration of simplex search and the MFO strategy. An adaptive experiment number (AEN) strategy is also presented to enhance the search method. To demonstrate the feasibility of the MFO method, it is used in the quality optimization of injection molding, a typical batch process with short cycle time and low operational cost. A comparison of the use of the MFO method with the MBO method whose models were developed with pure quadratic regression and Kriging methods is also presented. Experimental results demonstrate the efficient performance of the proposed method.